Name: Jay Goyal¶

Roll No: C017

Btech EXTC

Experiment-6

Aim:

Plot the Spectrogram

Importing the libraries

from scipy.io import wavfile
from IPython.display import Audio
import numpy as np
import matplotlib.pyplot as plt

fs1, data1 = wavfile.read(r'C:\Users\JAY GOYAL\audio of sp prac\exp_5\Forest.wav')
Audio(data=data1, rate=fs1)

print("sampling frequency: ",fs1)
plt.plot(data1)
plt.xlabel("n -->")
plt.ylabel("Amplitude -->")

sampling frequency:  44100

Text(0, 0.5, 'Amplitude -->')

from scipy import signal              
fs = 1000
t = np.arange(1000)/fs
signals = np.sin(2*np.pi*100*t)
plt.figure(figsize=(20,4))
plt.title("Signal A")
plt.plot(t,signals)
plt.xlabel("Frequency", color='r')
plt.ylabel("Amplitude", color='r')

Text(0, 0.5, 'Amplitude')

import scipy
from scipy.io import wavfile
import scipy.fftpack as fft
from scipy.signal import get_window
import IPython.display as ipd
import librosa

file_natural="Forest.wav"
signal, sr=librosa.load(file_natural)
print('Original sampling rate is',sr,'Hz')
signal, sr=librosa.load(file_natural,sr=16000)
print('The new sampling rate is set to',sr,'Hz')
print('The duration of the current audio sample is',len(signal)/sr,'s')

Original sampling rate is 22050 Hz
The new sampling rate is set to 16000 Hz
The duration of the current audio sample is 4.68 s

Displaying the waveplot

import librosa.display

librosa.display.waveplot(signal,sr=sr,color='purple')
plt.title('Time domain input')
plt.xlabel('Time')
plt.ylabel('Amplitude')
plt.show()

Normalising the audio

def normalize_audio(audio):
    audio = audio / np.max(np.abs(audio))
    return audio

signal = normalize_audio(signal)
plt.figure(figsize=(15,4))
plt.plot(np.linspace(0, len(signal) / sr, num=len(signal)), signal)
plt.grid(True)

Applying FFT

ft=np.fft.fft(signal)
ft=np.abs(ft)
freq=np.linspace(0,sr,len(signal))
plt.plot(freq,ft,color='orange')
plt.title('Output after applying inbuilt FFT function')
plt.xlabel('Frequency')
plt.ylabel('Amplitude')
plt.show()

Applying STFT

stft_n=1024
hop_len=256
stft=librosa.core.stft(signal,hop_length=hop_len,n_fft=stft_n)
specgram=librosa.amplitude_to_db(np.abs(stft))
librosa.display.specshow(specgram,sr=sr,hop_length=hop_len,cmap='magma')
plt.title('Spectogram using inbuilt function')
plt.xlabel('Time frame windows')
plt.ylabel('Frequency')
plt.colorbar()
plt.show()

fs2, data2 = wavfile.read('test.wav')
Audio(data=data2, rate=fs2)

print("sampling frequency: ",fs1)
plt.plot(data1)
plt.xlabel("n -->")
plt.ylabel("Amplitude -->")

sampling frequency:  44100

Text(0, 0.5, 'Amplitude -->')

from scipy import signal              
fs = 1000
t = np.arange(1000)/fs
signals = np.sin(2*np.pi*100*t)
plt.figure(figsize=(20,4))
plt.title("Signal A")
plt.plot(t,signals)
plt.xlabel("Frequency", color='r')
plt.ylabel("Amplitude", color='r')

Text(0, 0.5, 'Amplitude')

file_natural="test.wav"
signal, sr=librosa.load(file_natural)
print('Original sampling rate is',sr,'Hz')
signal, sr=librosa.load(file_natural,sr=16000)
print('The new sampling rate is set to',sr,'Hz')
print('The duration of the current audio sample is',len(signal)/sr,'s')

Original sampling rate is 22050 Hz
The new sampling rate is set to 16000 Hz
The duration of the current audio sample is 2.0 s

librosa.display.waveplot(signal,sr=sr,color='purple')
plt.title('Time domain input')
plt.xlabel('Time')
plt.ylabel('Amplitude')
plt.show()

def normalize_audio(audio):
    audio = audio / np.max(np.abs(audio))
    return audio

signal = normalize_audio(signal)
plt.figure(figsize=(15,4))
plt.plot(np.linspace(0, len(signal) / sr, num=len(signal)), signal)
plt.grid(True)

ft=np.fft.fft(signal)
ft=np.abs(ft)
freq=np.linspace(0,sr,len(signal))
plt.plot(freq,ft,color='orange')
plt.title('Output after applying inbuilt FFT function')
plt.xlabel('Frequency')
plt.ylabel('Amplitude')
plt.show()

stft_n=1024
hop_len=256
stft=librosa.core.stft(signal,hop_length=hop_len,n_fft=stft_n)
specgram_1=librosa.amplitude_to_db(np.abs(stft))
librosa.display.specshow(specgram_1,sr=sr,hop_length=hop_len,cmap='magma')
plt.title('Spectogram using inbuilt function')
plt.xlabel('Time frame windows')
plt.ylabel('Frequency')
plt.colorbar()
plt.show()

plt.figure(figsize=(10,16))
plt.subplot(5,1,1)
print("Sampling frequency of Forest.wav file:",fs1)
plt.title("Spectogram using inbuilt function for forest")
plt.plot(data1)
plt.xlabel("Time frame in windows")
plt.ylabel("Frequency")

plt.figure(figsize=(10,16))
plt.subplot(5,1,2)
print("Sampling frequency of test.wav file",fs2)
plt.title("Spectogram using inbuilt function for test")
plt.plot(data2)
plt.xlabel("Time frame in windows")
plt.ylabel("Frequency")

plt.figure(figsize=(10,16))
plt.subplot(5,1,3)
print("Sampling frequency of Forest.wav file",fs1)
plt.title("Spectogram using inbuilt function for Forest")
plt.plot(specgram)
plt.xlabel("Time frame in windows")
plt.ylabel("Frequency")

plt.figure(figsize=(10,16))
plt.subplot(5,1,4)
print("Sampling frequency of test.wav file",fs2)
plt.title("Spectogram using inbuilt function for test")
plt.plot(specgram_1)
plt.xlabel("Time frame in windows")
plt.ylabel("Frequency")
#plt.plot(specgram)
#plt.plot(specgram_1)

Sampling frequency of Forest.wav file: 44100
Sampling frequency of test.wav file 8000
Sampling frequency of Forest.wav file 44100
Sampling frequency of test.wav file 8000

Text(0, 0.5, 'Frequency')

import plotly.graph_objects as go

y = normalize_audio(fs1)
fig = go.Figure()
fig.add_trace(go.Scatter(
            y = data1,
            line =  dict(shape =  'spline' ),
            name = 'forest file'
            ))

fig = go.Figure()
y1 = normalize_audio(fs2)
fig.add_trace(go.Scatter(
            y = data2,
            line =  dict(shape =  'spline' ),
            name = 'test file'
            ))
fig.show()

Artificial Speech Signal

fs=22050
T1=1
T2=1
t1=np.arange(0,T1,1/fs)
t2=np.arange(0,T2,1/fs)
F1=1000
F2=2000

xx1=np.sin(2*np.pi*F1*t1)   # signal with freq=1kHz and amp=1
xx2=2*np.sin(2*np.pi*F2*t2) # signal with freq=2kHz and amp=2
xx=np.concatenate([xx1,xx2])
xx.shape

(44100,)

plt.figure(figsize=(15,5))
plt.plot(xx)
plt.title("Recorded Speech File",size=15)
plt.xlabel("n -->")
plt.ylabel("Amplitude -->")

Text(0, 0.5, 'Amplitude -->')

n_fft=256

fs

22050

FL=int(10*10**-3*fs)
FL

220

OV=0.25*FL
OV

55.0

FS=int(FL-OV)
FS

165

X1=librosa.stft(xx, n_fft=n_fft, hop_length=FS, win_length=FL)

plt.plot(abs(X1[:,10]))

[<matplotlib.lines.Line2D at 0x1691911bd00>]

S1=librosa.amplitude_to_db(abs(X1)) ##Log Spectrum
import librosa.display
plt.figure(figsize=(15,5))
librosa.display.specshow(S1,sr=fs,hop_length=FS,x_axis='time',y_axis='linear')
plt.colorbar(format='%+2.0f dB')

<matplotlib.colorbar.Colorbar at 0x16919234a60>

fig = go.Figure()
y = normalize_audio(xx)
fig.add_trace(go.Scatter(
            y = xx,
            line =  dict(shape =  'spline' ),
            name = 'artificial file'
            ))
fig.show()

Conclusion

So initially after importing the audio files we apllied fft to them, then normalised it and using librosa.display function in the forest.wav file the output was quite clear as comapred to the test.wav file which clearly depicts the the frequency range in decibels.Similar steps were followed for the artificial signal generated.